Image Fusion Technology Based On Multi-resolution Analysis

Image fusion is a process of combining different information from images of multiplesensors, then extract the complementary information to construct a fused image with abundantinformation. Image fusion can be applied in many field, it has broad prospects fordevelopment. Therefore, how to design an efficient image fusion algorithm has been put intothe agenda.In this paper, some of the important issues in multi-sensor image fusion have beendiscussed. The concrete work is shown as follows:We gave the concept of multi-sensor image fusion and the development of multi-sensorimage fusion technology, analysis characteristics of multi-sensor images in order to choose aproper fusion algorithm, we also gave the basic framework of the multi-sensor image fusionand image preprocessing method (geometric correction, image denoising, registration, etc),summarized the subjective and objective evaluation of the image fusion method. This articlefocused on analysis and summarizes the major multi-resolution image fusion technology andits features, we studied the pyramid method, wavelet transform, wavelet frame, Contourlettransform, Nonsubsampled Contourlet transform and the improved NonsubsampledContourlet transform. Analysis shortcomings and inadequacies of the above mentionedmulti-resolution method, several image fusion algorithms based on multi-resolution analysishave been put forward.The pyramid decomposition at different scales of the image contains redundantinformation. For the wavelet frame transformation, there is no sampling operation withtime-shift invariant properties can obtain relatively ideal fusion results, but the time efficiencyis not so good, therefore, in this paper, an effective image fusion algorithm based on the liftingscheme of wavelet transform has been proposed, The proposed algorithm is significantlybetter than the other algorithms in simulation time. The experiments proved that the proposedalgorithm is very effective in multi-focus image fusion.For the two-dimensional image processing, because of the limited directions, wavelettransform may lead to the fusion of image detail fuzzy, in order to make a better use of threedirections high-frequency wavelet sub-bands, we propose an image fusion algorithm based onthe structure direction information of lifting wavelet transform, fusion results shows that wecan get better result both in up-and-down focus, left-and-right focus and colorful imagefusion.Although Contourlet transform overcomes the limited directions of wavelet transform, but the Contourlet transform contains a sampling operation which may cause spectrumaliasing. Nonsubsampled contourlet transform (NSCT) with no sampling steps can achieve abetter result, but the use of pyramid transform in NSCT may cause the loss of subtleinformation. Therefore, we propose a nonseparable wavelet transform multi-scale directionanalysis (NSWMDA) combined with the pulse coupled neural network image fusionalgorithm. Two sets of simulation experiments have proved the effectiveness of the proposedalgorithm.